fix: close stale-decision races, honor user edges, normalize --at, refuse truncating exports

CHANGELOG.md

@@ -1,5 +1,51 @@
 # Changelog
 
+## 6.2.9 — 2026-07-10
+
+Independent re-audit of the 6.2.8 hardening release: one line-by-line
+auditor over the whole file, one differential regression hunter that
+first reproduced 6.2.7's defects to validate the 6.2.8 fixes, one
+claims-and-channels auditor, plus a full maintainer read of every file.
+All of 6.2.8's headline numbers reproduced exactly; the fixes below close
+what was still standing. Every fix was proven red with a regression test
+before the change:
+
+- **The last stale-decision races are closed.** The merge keeps a
+  per-node freshness baseline (last_accessed / access_count as of our
+  last disk read) and re-validates stale decisions against the CURRENT
+  disk copy inside the lock: a dream computed on a stale view can no
+  longer prune a memory another process confirmed meanwhile (the
+  GRACE_DAYS promise now holds under concurrency), a stale decayed
+  weight no longer min()s down a fresh confirm, and persisting a
+  duplicate-remember confidence upgrade no longer whole-copies stale
+  counters over a concurrent confirm — confidence merges max-wins as its
+  own field, the same way counters merge as deltas (6.2.8's way of
+  persisting it had traded one loss for another). Prune events are
+  journaled AFTER the save, and only for prunes that actually landed;
+  vetoed prunes are reported.
+- **The contradiction scan flags without clobbering.** An existing user
+  link between a conflicting pair keeps its relation and earned weight;
+  the conflict edge is only created where no user edge exists — "never
+  silently destroys data" covers edge metadata too.
+- **`recall --at` normalizes compact and tz-aware forms.** On 3.11+,
+  `--at 20260101` parsed but compared lexicographically against dashed
+  stamps ('-' < '0'), so every same-year fact looked valid at that past
+  date; all accepted forms now normalize to the naive dashed ISO the
+  graph stores before any comparison (older interpreters keep rejecting
+  compact forms as a usage error).
+- **A hand-deleted END guard no longer costs user content.** Export used
+  to rewrite the file and silently drop everything after BEGIN; the
+  damaged file is now left untouched and reported as skipped with the
+  reason.
+- **The duplicate-remember boost is one number.** The reopen path
+  persisted +0.2 while the plain path's delta replay persisted +0.15;
+  both are BOOST_PER_ACCESS now.
+- Known merge tradeoff, now stated: a long-lived stale process can
+  re-add a node another process pruned meanwhile (additive union favors
+  never losing writes; the next dream re-prunes it).
+- 220 tests. Mutation kill rate: 42% on the seeded 120-mutant sample.
+
+
 ## 6.2.8 — 2026-07-10
 
 Exhaustive line-by-line audit of code, tests, benchmarks, documentation,

README.ar.md

@@ -16,8 +16,8 @@
 كودكس، كرسور، زد)، و`CLAUDE.md` (كلاود كود)، و`GEMINI.md` (جيمناي) — وتُتبنّى تلقائيًّا في `.cursorrules` و`.windsurfrules` و`.clinerules` و`.roo/rules/mind.md` للمشاريع التي تستخدم تلك الأدوات أصلًا.
 
 ```bash
-curl -fsSLO https://raw.githubusercontent.com/Da7-Tech/mind/v6.2.8/mind.py
-python3 -c "import hashlib;h=hashlib.sha256(open('mind.py','rb').read()).hexdigest();assert h=='b3e944ce9103bd5e353e09a12f434721cb274bc919326361039104889c6c5e03',h;print('mind.py: OK')"
+curl -fsSLO https://raw.githubusercontent.com/Da7-Tech/mind/v6.2.9/mind.py
+python3 -c "import hashlib;h=hashlib.sha256(open('mind.py','rb').read()).hexdigest();assert h=='6dda66fcdb4352fb215f0f285d356e4a110617ff7714f3bdd1d96ab85f766663',h;print('mind.py: OK')"
 python3 mind.py init
 python3 mind.py remember "قاعدة بيانات المشروع بوستغرس ١٦"
 python3 mind.py recall "ما قاعدة البيانات"
@@ -52,7 +52,7 @@ python3 mind.py dream        # بين الجلسات: نسيان وتوطيد و
 
 **التمييز بالقياس** (`bench/discrim.py`، يعمل في CI): تدقيقان مستقلان أصابا حين قالا إن استرجاع الإبرة من ضجيج نظيف لا يقيس التفريق بين حقائق تتشارك المفردات — فهذا المقياس مشتِّتاته كلها متنافسة معجميًّا، وفيه حالات فشل التدقيقَين حرفيًّا («ما اسمي» ضد «اسم الملف يجب أن يطابق اسم الصنف» بالعربية والإنجليزية). النتيجة الحالية: **١٢/١٢** ببوابة ≥٠٫٨٥ في CI كي لا يتراجع التمييز صامتًا مجددًا.
 
-الاختبارات: **٢١٣ اختبارًا** بمكتبة بايثون القياسية وحدها + مشوّش (٤٢٠ حالة كاملة/١٦٠ في CI) + مطفّر بنسبة قتل **٣٩٪** على العينة المبذورة ذات ١٢٠ طفرة.
+الاختبارات: **٢٢٠ اختبارًا** بمكتبة بايثون القياسية وحدها + مشوّش (٤٢٠ حالة كاملة/١٦٠ في CI) + مطفّر بنسبة قتل **٤٢٪** على العينة المبذورة ذات ١٢٠ طفرة.
 
 ## موضعها بين الأدوات — وظائف مختلفة لا لوحة فائزين
 
@@ -153,7 +153,7 @@ python3 mind.py correct "قاعدة ماي إس كيو إل" "قاعدة الب
 ## التطوير
 
 ```bash
-python3 -m unittest discover -s tests   # ٢١٣ اختبارًا
+python3 -m unittest discover -s tests   # ٢٢٠ اختبارًا
 python3 bench/bench.py                  # أرقام العربية/الإنجليزية
 python3 bench/multilang.py              # ٨ لغات غير مضبوطة
 python3 bench/discrim.py                # التمييز بمشتتات متنافسة

README.md

@@ -23,8 +23,8 @@ adopted automatically by `.cursorrules`, `.windsurfrules`, `.clinerules`
 and `.roo/rules/mind.md` in projects that already use those tools.
 
 ```bash
-curl -fsSLO https://raw.githubusercontent.com/Da7-Tech/mind/v6.2.8/mind.py
-python3 -c "import hashlib;h=hashlib.sha256(open('mind.py','rb').read()).hexdigest();assert h=='b3e944ce9103bd5e353e09a12f434721cb274bc919326361039104889c6c5e03',h;print('mind.py: OK')"
+curl -fsSLO https://raw.githubusercontent.com/Da7-Tech/mind/v6.2.9/mind.py
+python3 -c "import hashlib;h=hashlib.sha256(open('mind.py','rb').read()).hexdigest();assert h=='6dda66fcdb4352fb215f0f285d356e4a110617ff7714f3bdd1d96ab85f766663',h;print('mind.py: OK')"
 python3 mind.py init
 python3 mind.py remember "the project database is postgres 16"
 python3 mind.py recall "which database do we use"
@@ -105,7 +105,7 @@ rounds had missed (see CHANGELOG 5.5.0).
 must catch them. Its first run exposed 17 behaviors the tests didn't
 actually pin down — each is now locked by a dedicated regression test
 (raw kill rate on the seeded 120-mutant sample: 33% at first run,
-**39%**
+**42%**
 on this release — the sample is re-drawn whenever the file changes, so
 the number moves a few points between releases; it is remeasured and
 republished each time because hiding it would be the exact sin this
@@ -115,7 +115,7 @@ local calculations superseded by the locked merge, platform-only branches,
 boundary/display constants, and ranking calibration guarded by the CI
 benchmark gates rather than exact unit assertions; the tool prints every one.
 
-Test suite: **213 tests**, stdlib `unittest`, `python3 -m unittest discover -s tests` —
+Test suite: **220 tests**, stdlib `unittest`, `python3 -m unittest discover -s tests` —
 including regression tests for concurrency (parallel writers must not lose
 each other's memories), destructive-op gating, corrupt-graph recovery, and
 a mutation-kill class where every test pins a behavior the suite
@@ -341,7 +341,7 @@ optional, since auto-dream already covers it.
 ## Development
 
 ```bash
-python3 -m unittest discover -s tests   # 213 tests
+python3 -m unittest discover -s tests   # 220 tests
 python3 bench/bench.py                  # reproduce the EN/AR numbers
 python3 bench/multilang.py              # 8 untuned languages
 python3 bench/soak.py                   # 180 simulated days

SECURITY.md

@@ -42,5 +42,7 @@ hardening suggestions. You can expect an initial response within a few days.
   store secrets, credentials, private personal data, or untrusted prompt text.
 - Node ids are `md5[:12]` content addresses — no security property is
   derived from them.
-- No network access, subprocess execution, or eval — the file can be fully
-  audited in one sitting (~2,900 lines).
+- No network access, no spawned processes, no eval — the file can be
+  fully audited in one sitting (~3,000 lines). (`subprocess` is imported
+  on Windows solely for its `list2cmdline` quoting helper; nothing is
+  ever executed.)

SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: mind
 description: Project memory graph with recall, provenance, and dreams.
-version: 6.2.8
+version: 6.2.9
 author: Da7 (Da7-Tech)
 license: MIT
 platforms: [linux, macos, windows]
@@ -36,7 +36,7 @@ built-in `memory` tool — and it is not a RAG system for large corpora.
 
 - `python3` (3.9+) and `curl` on PATH — nothing else: no API keys, no
   server, no packages. The tool is one stdlib-only file, MIT-licensed,
-  from https://github.com/Da7-Tech/mind (213 tests + benchmarks incl.
+  from https://github.com/Da7-Tech/mind (220 tests + benchmarks incl.
   10 languages + discrimination + fuzzer + 180-day soak test run in its CI
   on Linux/macOS/Windows).
 
@@ -47,8 +47,8 @@ tag and integrity-checked:
 
 ```bash
 cd <project>
-curl -fsSLO https://raw.githubusercontent.com/Da7-Tech/mind/v6.2.8/mind.py
-python3 -c "import hashlib;h=hashlib.sha256(open('mind.py','rb').read()).hexdigest();assert h=='b3e944ce9103bd5e353e09a12f434721cb274bc919326361039104889c6c5e03',h;print('mind.py: OK')"
+curl -fsSLO https://raw.githubusercontent.com/Da7-Tech/mind/v6.2.9/mind.py
+python3 -c "import hashlib;h=hashlib.sha256(open('mind.py','rb').read()).hexdigest();assert h=='6dda66fcdb4352fb215f0f285d356e4a110617ff7714f3bdd1d96ab85f766663',h;print('mind.py: OK')"
 python3 mind.py init
 ```
 
@@ -129,7 +129,7 @@ their rule files synced too (adopted only when present).
 ## Verification
 
 ```bash
-cd "$(mktemp -d)" && curl -fsSLO https://raw.githubusercontent.com/Da7-Tech/mind/v6.2.8/mind.py && python3 -c "import hashlib;h=hashlib.sha256(open('mind.py','rb').read()).hexdigest();assert h=='b3e944ce9103bd5e353e09a12f434721cb274bc919326361039104889c6c5e03',h;print('OK')" && python3 mind.py init >/dev/null && python3 mind.py remember "the sky signal is 7413" >/dev/null && python3 mind.py recall "sky signal"
+cd "$(mktemp -d)" && curl -fsSLO https://raw.githubusercontent.com/Da7-Tech/mind/v6.2.9/mind.py && python3 -c "import hashlib;h=hashlib.sha256(open('mind.py','rb').read()).hexdigest();assert h=='6dda66fcdb4352fb215f0f285d356e4a110617ff7714f3bdd1d96ab85f766663',h;print('OK')" && python3 mind.py init >/dev/null && python3 mind.py remember "the sky signal is 7413" >/dev/null && python3 mind.py recall "sky signal"
 ```
 
 Expected: one result containing `7413` with a printed memory id.